Simultaneous inversion and geobody extraction to estimate carbon storage potential in the Ivishak Formation of the Alaska North Slope, United States

Alaska has been a prolific oil-producing state since the 1980s when oil first ran through the Trans-Alaska Pipeline System. More than 18 billion barrels have been produced in the state, and another 5 billion are estimated to be recoverable (Masterson and Holba, 2021). However, two strong trends presage changes in Alaska's energy industry: oil production has steadily declined, and demand for low carbon emission fuels is rising. A tool that can help address these issues is carbon storage. Carbon storage is a methodology that captures and injects CO₂ underground, which can greatly offset the emissions from produced hydrocarbons as well as use existing infrastructure (pipeline, roads, gravel pads, and surface facilities) to extend its value. The geology of the North Slope of Alaska is favorable for carbon capture. Many legacy oil fields contain trends of high-quality sands that are water wet outside the hydrocarbon accumulation. However, the supply of CO₂ for injection is presently limited. The North Slope carbon emission point sources are due largely to combustion (Patil et al., 2006; Shellenbaum and Clough, 2010). Carbon storage demand could rise in the event of a natural gas pipeline being constructed. Natural gas supporting the line would contain a large amount of CO₂ that is mixed in the gas stream that would need to be separated and sequestered. This could provide a significant source of CO₂ and increase carbon capture, utilization, and storage (CCUS) demand.

In anticipation of future pipelines and repurposing declining oil fields, there is a need to properly characterize carbon storage targets near infrastructure on Alaska's North Slope. One legacy oil field that provides an excellent opportunity is the Kuparuk oil field. Kuparuk is an Upper Jurassic–Lower Cretaceous oil field where operators target highly productive sands in both structural and stratigraphic traps (Carman and Hardwick, 1983; Masterson and Eggert, 1992). Directly below this prolific field sits a deeper Triassic sand known as the Ivishak Formation. Immediately to the east, a structural trap exists where the Ivishak is a principal reservoir for the Prudhoe Bay oil field. The numerous well penetrations through the field show that gross thicknesses can reach more than 650 ft (approximately 200 m), and the field average for porosity is 22% (Woodhouse, 1998; Masterson and Holba, 2021). However, at the Kuparuk area, there are no hydrocarbon traps in the Ivishak Formation, and the handful of wells that reach the deep Ivishak reservoir are water wet. There is a risk that wells penetrating the Ivishak would have wellbore integrity issues and possibly leak at the surface. However, there are relatively few wells that reach the Ivishak interval, and these could likely be plugged or utilized with minimal cost. Additionally, the reservoir is normally pressured and would allow for high rates of fluid injection. The high reservoir quality sands, few wellbore penetrations, and normally pressured aquifer are conditions that make the Ivishak Formation a high-potential CCUS target.

Little work exists showing storage estimates for site-specific locations on the North Slope. Basin-wide studies have previously recognized the potential for large quantities of carbon storage in Alaska (Shellenbaum and Clough, 2010; Craddock et al., 2014), but a detailed analysis using seismic data and well-log information is lacking. A more detailed reservoir characterization of injection sites near existing infrastructure would provide an understanding for initial strategy and planning for future possible CCUS projects. This present study leverages seismic inversion to map and extend petrophysical parameters across the core area of the Kuparuk River oil field. Prestack inversion outputs correlate to petrophysical parameters, which are key to understanding where higher-quality reservoir exists. The reservoir extent can be defined by using a 3D spatial visualization technique called geobody extraction. Geobody extractions from these inversion outputs provide an estimate of pore space available for carbon storage. The results show that inversion can improve mapping interpretations by providing a stable inversion response to pick and by 3D visualizations to provide better estimates for carbon storage potential. Operators on the North Slope could use such a methodology to develop a new perspective on storage availability near infrastructure.

This study focuses on an area near existing infrastructure that is overlain by a 3D seismic survey (Figure 1). The primary data used in the study are 3D seismic survey and well logs from eight wells. The 3D seismic survey is a merged volume covering 930 square miles (2400 km²). Eight smaller subvolumes were combined to form the merged survey. Nominal fold varies from 35 to 110 across the subvolumes. During reprocessing, the surveys were 5D interpolated, which artificially boosted fold values to more than 200. However, the processing objective was to preserve prestack amplitudes and minimize noise. Well-log data sets containing resistivity, bulk density, sonic, and gamma ray measurements are available in all the wells (Table 1). The Ivishak Formation has measured log data across the entire formation by these well logs, but deeper zones have variable coverage. Petrophysical interpretations show high water saturations in all the wells with no evidence for hydrocarbon accumulations. Of the eight wells within the study area, the 2F-20 is the only well containing a quality shear sonic log for the Ivishak interval (Figure 1).

Four megasequences make up the stratigraphy on the North Slope: Franklinian, Ellesmerian, Beufortian, and Brookian (Hubbard, 1988) (Figure 2). This study focuses on the Ellesmerian Sequence. The Ellesmerian Sequence corresponds to the Mississippian until the Early Jurassic and ended with the opening of the Canada Basin as the Alaskan and Canadian margins rifted (Moore et al., 1994). During this time, the Arctic Alaska Basin was part of a passive margin along the Laurentian continental margin, with sediments that were shed from the northern highlands of the Arctic Platform and thinning southward (Houseknecht, 2019). Deposition in the Ellesmerian begins with the Endicott Group, which consists of conglomerates, sands, and shales deposited directly onto the metamorphosed Franklinian Basement (Figure 2). After this early deposition, the continental margin was flooded, and the carbonate Lisburne Group deposited. This formation is a key seismic marker for this analysis and the basal pick for the seismic inversion. The Lisburne Group is mainly limestone and dolomite with minor amounts of shale, sandstone, and evaporites (Dumoulin et al., 2013). Early explorers expected this formation to be the primary hydrocarbon reservoir based on outcrops where they observed coarse carbonate grains and porosity (Masterson and Holba, 2021). On the contrary, porosity at reservoir conditions is primarily driven by dolomitization and is difficult to predict. The change in seismic velocity and bulk density between the overlying Sadlerochit Group and the Lisburne makes the latter a key horizon for inversion analysis.

The Sadlerochit is next going up the stratigraphic record and represents an overall transgression along the continental margin with highlands northward and open ocean to the south (Figure 2). Within this group is the Ivishak Sandstone, which is the main formation of interest. In the Early Triassic, the Ivishak was deposited by prolific fluvial-deltaic systems that created multistoried channels varying between meandering and braided. During the Prudhoe Bay field development, the zone was mapped with a lithostratigraphic framework where four different petrophysical zones divide the formation (Wadman et al., 1979). On a regional scale, the Ivishak can range in thickness from 50 ft to more than 600 ft (15–180 m) (Masterson and Holba, 2021). At the close of the Ellesmerian Sequence, a regional transgressive cycle results in the deposition of the Shublik Formation and the Sag River Sandstone. The Shublik is a prolific source rock for the basin and, specifically for Prudhoe Bay, can also act as a seal. The Sag River is a relatively thin sandstone reservoir that is widespread across the North Slope and often contains hydrocarbon accumulations when a structural trap is present.

The Beaufortian marks the end of passive margin deposition and the beginning of rift-related deposition (Montgomery, 1998). Tectonic events during this time created the Barrow Arch, which is a critical structure for many of the hydrocarbon accumulations on the North Slope, including the Kuparuk oil field, which is the focus area of this work. At the close of the Beaufortian Sequence, a regionally extensive shale known as the high radioactive zone (HRZ) is deposited (Figure 2). The HRZ marks a relative sea-level rise and is the shallowest marker used for mapping and correlation in this study.

Data conditioning. First, the well-log data were reviewed and tied to the 3D seismic data. Well ties showed a consistent correlation throughout the eight control wells. A statistical, zero-phased wavelet provided stable well ties (Figure 3). The seismic data set used for initial well ties was the near-angle partial stack (0°–20°). The HRZ, Sag River, and Lisburne are key reflectors tying seismic data to the logs. Both the Sag River and Lisburne have strong contrast in compressional sonic velocity and bulk density when compared to the overburden (Figure 3). This response is consistent to map across the seismic volumes. At these events, the logs tie consistently across the eight wells used in the inversion. Shear sonic logs are necessary in the inversion to derive the shear-wave velocity. To remedy the lack of shear sonic logs, the 2F-20 well log was used as the training data set for a discrete neural network (DNN), which predicted and calculated shear sonic logs in the remaining wells. DNN is a regression method employed to predict a single output target (shear sonic). Inputs for the predicted shear sonic log were compressional sonic velocity and bulk density.

Second, key seismic reflectors were identified and mapped in the 3D seismic volumes. Consistent seismic markers observed in the well-tie analysis gave confidence in the mapping process. The top horizon used was the HRZ, which is an Early Cretaceous shale bed in the Brookian Sequence (Figure 2). The lower surface picked was the top of the Lisburne Group. The increase in seismic velocity provided a consistent marker across the survey. Amplitudes at the Ivishak level were noisy and challenging to interpret consistently. As an alternative, the Sag River sandstone, a nearby reliable marker, was chosen.

Third, common midpoint gathers were reviewed to ensure proper flattening and determine the level of noise contamination in the data. The gathers showed that amplitudes beyond 40° were noisy and would contaminate the signal. Between 0° and 40°, the Ivishak level, as well as the corresponding overburden and underburden, showed good-quality amplitude response. Initially, gathers were in the offset domain from the Greater Kuparuk Area seismic merge (Figure 1). Due to the 5D interpolation during processing, these gathers are high-fold, with more than 200 traces per bin. For understanding the amplitude variation with offset (AVO), viewing gathers in the angle domain can be easier to review and communicate. To make this change, migration velocities were used to convert the gathers from offset to angle. During this step, the fold was reduced to 16 without losing data fidelity. The high-fold data were difficult to use due to size constraints in the software packages. In addition, reducing the number of traces made the data easier to manage and greatly reduced the processing time for inversion. Sixteen traces were more than sufficient to properly measure the AVO. This data conditioning and review gave confidence that the seismic inversion would provide key reservoir information to aid the estimation of carbon storage potential.

Inversion and neural networks. The deterministic seismic inversion requires a low-frequency model, which is created from the well data and horizons described earlier. This model provides a low-frequency background impedance trend that is added to the relative impedance derived from the seismic data. Due to the gathers being good quality out to 40°, this study implements a simultaneous inversion where the gather information will be inverted to provide V_P, V_S, and density outputs. These inversion outputs can be combined into useful seismic attributes that inform the lithology and fluid variations within the zone of interest. The inversion was executed for the full area, with a vertical range from 100 ms above the Sag River down to 200 ms below the Lisburne horizon (Figure 4).

Variations in V_P, V_S, and density can be useful for understanding subsurface lithologic and fluid variations (Castagna, 1993). Additionally, impedance contrasts can be exploited to describe subsurface variations (Fatti et al., 1994). For this study, impedances and Lamé parameters are derived to define potential carbon storage. There are two Lamé parameters, λ and μ, which are shown in equations 1 and 2. The parameters arrange velocity and density information into moduli informing on the rigidity and incompressibility for a given rock unit (Goodway et al., 1997; Dufour et al., 2002). These parameters can be more sensitive to pore fluids and lithology variations, which are necessary outputs for evaluating reservoirs for carbon storage potential.

where ρ is density, V_P is P-wave velocity, and V_S is S-wave velocity.

Seismic attributes can next be leveraged to extend the key petrophysical outputs away from the wellbores. This is done through multilinear regression analysis and neural networks (Hampson et al., 2001). This work implements neural networks due to the algorithm's ability to predict even in noisy data where there is a nonlinear relationship between inputs (Sahoo and Jha, 2017). For inversion, this work follows a workflow shown by Herrera et al. (2006). The first step uses multilinear regression to identify which seismic attributes are most useful for predicting the desired petrophysical parameter. These high-graded attributes are used as input in the neural networks and trained with wells containing porosity and Vshale logs. To avoid overtraining, the analysis ran validation tests that removed control wells and conducted a blind analysis. Minimized error during blind well testing is the goal to achieve a model that is not overtrained. Next, the trained relationship is applied to extend the petrophysical parameter across the seismic coverage — essentially placing a porosity and Vshale well log at each seismic sampled position. This output can now be interpreted at the reservoir level to highlight areas for greater CO₂ storage potential.

Mapping and neural network predictions. The simultaneous inversion outputs acoustic impedance, shear impedance, density, and V_P/V_S volumes. It is observed that density has the most error, which is driven by gathers cut at 40° angles due to data quality. Additionally, Poisson's ratio, λρ, and μρ were computed. Measured well-log data in the 2F-20 well showed that lambda rho and acoustic impedance can separate the high-porosity sands with lower Vshale values (Figure 5). Both attributes are heavily impacted by the acoustic impedance component, which has a greater impact on the near-angle amplitudes during inversion. Due to amplitudes being cut at 40°, the lack of far amplitudes likely explains why other attributes did not have strong correlations to porosity and Vshale.

Mapping key horizons was greatly improved using the relative inverted volumes (Figure 6). Inversion removes wavelet effects and now provides a stable zero crossing at the Sag River that can largely be autotracked across the area. Previously, the Ivishak and Echooka horizons could not be interpreted consistently on the amplitude data. Relative inversion volumes remove the background trend imposed from well data, thus allowing the interpreter to only see impedance contrasts driven by the seismic data. Therefore, the relative inversion volumes were used extensively for mapping. The relative lambda rho attribute clearly delineated new zero-crossing boundaries that tie consistently to the Ivishak and Echooka tops. Improved mapping results allow for better constrained extractions that furnish Ivishak sand distribution and variation.

Neural networks allowed for prediction of high-quality sands away from the wellbore locations in terms of porosity and Vshale. The machine learning algorithm used a combination of inversion, calculated, bandpass, and AVO attributes for prediction. Application of the neural net relationship produced a high correlation between the predicted and modeled values for porosity and Vshale (porosity = 85%, Vshale = 89%) (Figures 7 and 8). These correlation percentages are produced by using the training wells to compare the measured and modeled logs. During the blind well testing, each well individually is left out of the analysis and the other wells are used to predict the values. The resulting correlation between the blind prediction and the measured data is called the validation correlation, which for porosity and Vshale were 62% and 68%, respectively. Neural net results were extrapolated across the area to create 3D volumes and allowed for geobody extractions. Geobodies proved to be an excellent tool for visualization and 3D manipulation of the data (Figure 9). Porosity and Vshale cutoffs for the extractions were porosity greater than 15% and Vshale less than 35%. Extractions show that the Ivishak sands are extensive and practically cover the entire area, with an average thickness for the area of 445 ft (136 m) (Figure 10). Mapping of the Vshale and porosity volumes implies a thick, laterally continuous fluvial-deltaic environment that contains well-connected, coalescing sand distribution. Local areas with greater thicknesses are possibly due to multistory channels or more available accommodation during deposition. However, seismic resolution is lower at this interval and individual depositional elements are not clearly defined. Chiefly, no major shale beds or large faults are observed in the mapping and extractions. The lack of major faulting and geobody extractions showing continuous sand packages suggest that large volumes of CO₂ could be injected with minimal risk of geologic compartmentalization.

Ivishak carbon storage estimation. Pressure and temperature trends inform that the study area is in an ideal depth for CO₂ injection. The minimum depth to reach supercritical CO₂ phase is approximately 3600 ft (1100 m), and the average depth to the Ivishak is 8750 ft (2670 m). Additionally, there is no significant overpressure event within the zone to impede fluid injection. Carbon storage potential is calculated using the Goodman equation for a brine-filled reservoir (equation 3) (Goodman et al., 2011):

where G_CO2 is geologic storage; A_t is total area; h_net is the net height of the reservoir, where Vshale is less than 35% and porosity is greater than 15%; ϕ is total porosity; ρ_CO2 is density of CO₂ at storage conditions; and E_saline is a CO₂ storage efficiency factor.

The extracted geobodies show high-quality sands (porosity greater than 15% and Vshale less than 35%) distributed throughout the study area, which is 560 square miles (1450 km²) and the average thickness is 445 feet (136 m). Porosity data from the wells show a zone average of 16%. Temperature and pressure data from the wells show CO₂ density, at reservoir conditions, to be 0.65 ml. For a CO₂ efficiency factor, the calculation uses the site-specific recommendation where P₁₀, P₅₀, and P₉₀ are 0.0462, 0.0679, and 0.1492, respectively (Goodman et al., 2011). For the targeted study area near-surface infrastructure, the Ivishak Formation holds P₁₀, P₅₀, and P₉₀ carbon storage potential of 1.19, 1.75, and 3.84 GT, respectively. Previous literature identified that the North Slope had potential for CO₂ storage at a large basin scale. Now, these results further support large geologic capacity with a refined CO₂ geologic storage estimate in the Kuparuk River Unit. This new characterization can aid decision makers in determining future CCUS strategy in Alaska.

Future work that could advance this analysis would start with plume modeling to better understand the CO₂ migration extent. A large uncertainty is involved with the efficiency factors that were used for the research, and better estimates for that uncertainty could greatly adjust estimates for storage in the Ivishak. Along with future modeling, further work is needed to develop a monitoring strategy for CCUS. Long-term storage would require a comprehensive monitoring program at the subsurface, surface, and atmospheric levels. Finally, the current work did not identify significant faulting in the Ivishak; however, faults do exist. More analysis should be conducted to understand the risk of leakage from faulting and the overburden seal capacity.

Key conclusions for this work can be summarized as follows:

• Simultaneous inversion aids greatly in mapping the top of the Ivishak and Echooka formations. Petrophysical properties can be predicted from seismic attributes using supervised neural networks.

• Geobody extractions are very helpful in mapping and visualizing the extent of Ivishak sands. For the core area, the sands average more than 400 ft (120 m) thick. No significant fault or compartmentalization is observed in these extractions.

• The Ivishak Formation is indeed a prolific sandstone across the Kuparuk River Unit. Estimates show between 1 and nearly 4 GT of CO₂ storage in the study area. This estimate is based on supervised neural network results trained from seismic attributes and well logs.

We appreciate ConocoPhillips for making seismic data available for use in this research. Log data used for this study were provided by TGS and the Alaska Oil and Gas Conservation Commission.

Well-log data are publicly available through the Alaska Oil and Gas Conservation Commission. The 3D seismic data are confidential and cannot be released.